Steel and Composite Structures, Vol. 31, No.

5 (2019) 427-435
DOI: https://doi.org/10.12989/scs.2019.31.5.427 427

Moment-rotation estimation of steel rack connection

using extreme learning machine
Mahdi Shariati 1,2a, Nguyen Thoi Trung 1,2b, Karzan Wakil 3c,

Peyman Mehrabi 4d, Maryam Safa 5 and Majid Khorami 6e
1
Division of Computational Mathematics and Engineering, Institute for Computational Science,
Ton Duc Thang University, Ho Chi Minh City, Vietnam
2
Faculty of Civil Engineering, Ton Duc Thang University, Ho Chi Minh City, Vietnam
3
Research Center, Sulaimani Polytechnic University, Sulaimani 46001, Kurdistan Region, Iraq
4
Department of Civil Engineering, K.N. Toosi University of Technology, Tehran, Iran
5
Institute of Research and Development, Duy Tan University, Da Nang 550000, Vietnam
6
Universidad UTE, Facultad de Arquitectura y Urbanismo, Calle Rumipamba s/n y Bourgeois, Quito, Ecuador

(Received March 6, 2018, Revised February 11, 2019, Accepted May 7, 2019)

Abstract. The estimation of moment and rotation in steel rack connections could be significantly helpful parameters for
designers and constructors in the initial designing and construction phases. Accordingly, Extreme Learning Machine (ELM) has
been optimized to estimate the moment and rotation in steel rack connection based on variable input characteristics as beam
depth, column thickness, connector depth, moment and loading. The prediction and estimating of ELM has been juxtaposed
with genetic programming (GP) and artificial neural networks (ANNs) methods. Test outcomes have indicated a surpass in
accuracy predicting and the capability of generalization in ELM approach than GP or ANN. Therefore, the application of ELM
has been basically promised as an alternative way to estimate the moment and rotation of steel rack connection. Further
particulars are presented in details in results and discussion.
Keywords: steel racks; moment rotation behavior; upright column; beam-end connector; ELM

1. Introduction and juxtaposed with the higher storage items‘ volume

(Shariati et al. 2010, 2012, 2015, 2017, 2018, Shah et al.
Estimation for behavior of structural members is useful 2015, 2016a, b, c, Chen et al. 2019).
for design of structures. This phenomenon has been used in Three dimensional structures have provided a direct and
many recent researches including steel-concrete beams, availability to the whole stored items and have been easily
reinforced concrete columns, etc. (Vo-Duy et al. 2017, demounted with a reassembling capability. The effectual
2018, Heydari and Shariati 2018, Ho-Huu et al. 2018, application of racks has required a material flexibility to
Hosseinpour et al. 2018, Ismail et al. 2018, Nasrollahi et al. constitute the racks. Respectively, cold-formed steel has
2018, Paknahad et al. 2018, Wei et al. 2018, Zandi et al. been regarded to the manufacture of these weird structures
2018, Abedini et al. 2019, Davoodnabi et al. 2019, Luo et (Kheyroddin et al. 2008, Gilbert and Rasmussen 2009,
al. 2019, Sajedi and Shariati 2019, Xie et al. 2019), and Bazzaz et al. 2011, 2012, 2014, 2015a, b, Gilbert et al.
may be useful to give design perspectives for utilizing 2012, Andalib et al. 2014, Shah et al. 2016b, Abedini et al.
newly developed beams and columns made of materials 2017, Anicic et al. 2017, Khorami et al. 2017b, Andalib et
with inherent tailored micro-structures (known as meta- al. 2018, Bazzaz et al. 2018, Paknahad et al. 2018) to allow
materials). Knowing the way of estimation of structural a hand-adjusting and rack parameters‘ resembling when
member behavior could help for more optimized and needed and to its good strength to weight rate. In contrast,
affordable design of them. Steel rack has been taken as an AS4084 has permitted the use of hot roll steel to be
applicable resolution to the situations by providing an manufactured, if rack should support the severe loading.
adequate and accessible storage, if low space is provided The current research has evaluated the improvement in
beam-column connection function under static load by
changing the most effectual characteristics. 32 different
Corresponding author, Mrs., tests by using a double cantilever testing method have been
E-mail: maryamsafa@duytan.edu.vn performed on the rack connection by changing beam depth,
a
Ph.D., E-mail: shariati@tdtu.edu.vn column thickness and beam end connector‘s depth. The
b moment-rotation and load-strain behavior of connections
Ph.D., E-mail: nguyenthoitrung@tdtu.edu.vn
c have been checked, also the effect of the elements on
Ph.D., E-mail: karzanwakil@gmail.com
d general connection function has been experimented. On the
M.Sc., E-mail: p.mehrabi@email.kntu.ac.ir
e other hand, a comparative research to gain the joint stiffness
Ph.D., E-mail: khoramimajid@yahoo.com
by using three various models as (1) the primary stiffness
Copyright © 2019 Techno-Press Ltd.
http://www.techno-press.org/?journal=scs&subpage=6 ISSN: 1229-9367 (Print), 1598-6233 (Online)
428 Mahdi Shariati, Nguyen Thoi Trung, Karzan Wakil, Peyman Mehrabi, Maryam Safa and Majid Khorami

model; (2) slope to half ultimate model; and (3) the equal model and portal frame testing model. Later, cantilever
area model is performed. testing has been extended by the attachment of one extra
The steel rack connection analysis has required a precise beam to other side of column, mightily named as double
on line determination of moment and rotation, accordingly, cantilever testing model. Although, it is difficult to equally
the estimation model of steel rack connection moment and load the connections on both right and left side of the
rotation through the Extreme Learning Machine (ELM) are column and the consequent deflection may not be the same,
used. Nowadays, computational models‘ applications to the related literature has proved that double cantilever test is
solve real shortcomings and determining the optimum more beneficial with better results than cantilever test. The
variables and performances have enormously been beam-end connector adjusted to moment, axial pull, and
considered in various scientific realms (Hamidian et al. shear, like to real frame, has mightily provided a
2012, Toghroli et al. 2014, Aghakhani et al. 2015, satisfactory estimating of shear and moment rate. The shear
Mohammadhassani et al. 2015, Toghroli 2015, Mansouri et to moment ratio in a real rack frame might be better
al. 2016, 2017, Safa et al. 2016, Toghroli et al. 2016, reproduced by double cantilever test. Consequently, in this
Khorami et al. 2017a, Sadeghi Chahnasir et al. 2018, study, the results of double cantilever test method have been
Sedghi et al. 2018, Shariat et al. 2018, Toghroli et al. 2018, used as testing arrangement. A brief report of this research
Zandi et al. 2018). Neural network (NN) would be recently is brought below.
nominated in various engineering era to solve the
complicated non-linear shortcomings by using classic 2.2 Specimen and material
parametric methods. On the other hand, back propagation
(BP), support vector machine (SVM), and hidden Markov Two types of beam-to-upright connections have been
model (HMM) are taken as training NN algorithms. The supplied; however, only one type has been experimented.
neural network‘s problem is in its learning time requirement Schematic diagrams of selected connection for beam-end
(Singh et al. 2008, 2012, Mansouri et al. 2016, Safa et al. connector are depicted in Fig. 1.
2016, Toghroli et al. 2016). an algorithm for single layer
feed forward NN as ELM specified to solve shortcomings 2.3 Material properties
occurred by gradient descent algorithm(s) like BP used in
ANNs has been nominated by Huang et al. (2004). ELM Cold-formed steel section(s) have been applied for
has also the capability of reducing the required training time column and beam. Also, the beam-end connectors are built
for one NN, in other word, using ELM has tremendously of hot roll steel. The material features of members and
raised the speed of learning process and generated a beam-end connectors are given in Table 1.
robust performance (Huang et al. 2006c). Thus, likewise
other algorithms, ELM is also used in diverse scientific
problem-solving fields (Mohammadhassani et al. 2013,
2014, Wang and Han 2014, Yu et al. 2014, Anicic et al.
2017), but a powerful algorithm with fast learning speed
and well function compared to the classic ones as BP. ELM
has focused to gain the lowest train error and weights‘
norm. In the current study, a predictive method of steel rack
connection moment and rotation has been created by ELM.
The results have indicated the capability of proposed model
in the prediction of energy usage in buildings. Also, the
results of ELM have been juxtaposed with the results of GP
and ANN. There is an attempt to retrieve the correlation
among the column-thickness, beam-depth, connector-
depth, load and steel rack connection moment and rotation.
The system is for predicting the moment and rotation of the
steel rack connection in terms of four input parameters.

2. Methodology Fig.1 Detail of the beam end connector (Shariati et al. 2018)

2.1 Testing

To study the behavior of beam-end connector, a Table 1 Specimens‘ material property (Shariati et al. 2018)
comprehensive research has been conducted recently Young‘s Poisson‘s Yield Ultimate
(Shariati et al. 2018). In this research design standards for Member
modulus (E) ratio (ν) strength (fy) strength (fu)
storage rack design including Rack Manufacturers Institute Column 280 382
(RMI 1997), Australian Standard (AS 4084) (2012), Storage
Equipment Manufacturing Association (SEMA 1985) and Beam 327 435
205 0.3
F´ed´eration Europ´eenne de la Manutention are the Beam-end-
304 353
recommend alternative testing methods as cantilever testing Connector
 Moment-rotation estimation of steel rack connection using extreme learning machine 429

Table 2 Beam dimensions 2.7 Input and output variables

Type of Width ‗b‘ Depth ‗h‘ Thickness ‗t‘
Four input parameters have been obtained from the
beam (mm) (mm) (mm)
experimental test results to analyze the steel rack
B1 40 92 1.5 connections moment as: column-thickness, beam-depth,
B2 40 110 1.5 connector depth and loading (Table 3). On the other hand, 5
B3 50 125 1.5 input parameters have been obtained to analyze the steel
rack connection rotation as column-thickness, beam- depth,
B4 50 150 1.5
connector depth, load and 5) moment (Table 4).

2.8 ELM
2.4 Specimen details ELM has been nominated for the single layer feed
forward NN (SLFN) architecture (Annema et al. 1994,
Totally, 32 experiments have been conducted composed Huang et al. 2006b), randomly choosing the input weights
of 4 trials of each specimen‘s set defined by 2 various and analytically determining the output weights of SLFN.
column-thickness, 4 various beam-depth and 2 different ELMS with its fast learn speed has no required many
umbers of tabs in beam-end connector. The column human interventions compared to conventional algorithm(s).
specimens have been spotted by their thicknesses. Column ELM has also the capability of analytically defining all
A is 2.0 mm thickness and column B is 2.6 mm thickness. network elements to prevent the trivial human interventions
Welding of beam in compression zone has resulted to a known as an influential algorithm with more merits like
weak point in the exposed portion of beam-end connector smooth use, fast learning, and high function, suitable for
with tabs to make the connector susceptible to tensile more non-linear activation and kernel performances.
failure and to reduce the final moment carrying the capacity
of connection (Abedini et al. 2019). Failure assessments in 2.8.1 SLFN
design have drawn many attentions (Bobaru et al. 2018, SLFN performance with L hidden nodes has been
Mehrmashhadi et al. 2019a, b). Down welding of the beam offered as mathematical explanation of SLFN, interlocking
has minimized the tensile failure of beam-end connector. the additive and RBF hidden-nodes (Huang et al. 2006a,
The cross section of the box beam and upright are Liang et al. 2006)
represented in Figs. 2(a) and (b). The beam sections‘ details
have been provided in Table 2. L
f L  x   iG  ai ,bi ,x  , x  Rn , ai  R n (1)
i 1
2.5 Test setting up
ai and bi = learning characteristics of hidden-nodes
Double cantilever testing has been followed for βi = the weight connecting ith hidden node to the output
predicting the moment-rotation behavior of the node
connection(s). Therefore, test has restrained the shear G (ai, bi, x) = output value of ith hidden node regarded to
displacement in the column(s) to be treated as a rigid body. input x
The column has been initially correlated and vertically
levelled below the load tool. Later, two beams are attached
to the right-left side of the column at its center. Small Table 3 Input parameters for steel rack connection moment
amount of pre-loading has been primarily used and the prediction
displacement measuring tools are adjusted. The schematic
representation of test setting up is illustrated in Fig. 3. Inputs Parameters description Parameters characterization
Input 1 Thickness of column (mm) Min: 2 Max: 2.6
2.6 Instrumentation Input 2 Depth of beam (mm) Min: 92 Max: 150
Input 3 Depth of connector (mm) Min: 205 Max: 250
Within the tests besides loading, three various
measurements have been performed to gain a data-set on Input 4 Loading (kN) Min: -23.184 Max: 4.218
connection behavior throughout the whole ranges of used
loading including: (1) performing of strain reading to steel-
yielding monitoring; (2) displacement measurement to gain Table 4 Input parameters for steel rack connection rotation
the load deflection behavior; and (3) rotation measurement prediction
to gain moment rotation properties. Inputs Parameters description Parameters characterization
Deflection measurements are performed at the selected
zone by linear variable differential transformer (LVDT) in a Input 1 Thickness of column (mm) Min: 2 Max: 2.6
measurement range (50-200 mm). Two digital inclinometers Input 2 Depth of beam (mm) Min: 92 Max: 150
are installed on both sides of the beam to directly gain the Input 3 Depth of connector (mm) Min: 205 Max: 250
beam-rotation (degree) by taking them on the beam sides. Input 4 Load (kN) Min: -23.184 Max: 4.218
Input 5 Moment (kNmm) Min: -0.0075 Max: 5.224372
430 Mahdi Shariati, Nguyen Thoi Trung, Karzan Wakil, Peyman Mehrabi, Maryam Safa and Majid Khorami

(a) Upright cross section (b) Beam cross section

Fig. 2 Beam and column section details (Shariati et al. 2018)

The additive hidden node with the activating perfor-

mance of g(x): R → R (say, sigmoid and threshold), G (ai, ACTUATOR
TESTING FRAME
bi, x) is

G  ai ,bi ,x   g  ai .x  bi  , bi  R INCLINOMETER COLUMN

(2) BEAM END CONNECTOR

BEAM BEAM
ai = vector- weight connecting the input layer to ith ROLLER SUPPORT
D1 D3
HINGE SUPPORT
hidden node L/4 L/4 L/4 L/4
bi = bias of ith hidden node ai DUAL GUAGE D2
`
x = inner product of vector ai and x in Rn PEDESTAL PEDESTAL

G (ai, bi, x) has been obtained for RBF hidden node with Fig. 3 Double cantilever test set-up diagram
activating function g(x): R → R (say, Gaussian), G (ai, bi, x) (Shariati et al. 2018)
as
G  ai ,bi ,x   g  bi x  ai  , bi  R  (3)
H  T (5)
th
ai and bi = center and impact factor of i RBF node Where
R+ = set of all positive actual variables
G  a1 ,b1 ,x1  G  aL ,bL ,x1  
 
RBF network is a specific SLFN with RBF nodes in its H( a,b,x )    (6)
hidden layer. For N arbitrary distinct samples   a1 ,b1 ,xN  G  aL ,bL ,xN   N L
 xi ,ti   Rn  Rm , xi is n×1 input vector and ti is m×1 target
vector. If one SLFN with L hidden nodes could approximate With
the N samples with zero error, then it means that there is βi, a  a1 ,,aL ; b  b1 ,,bL ; x  x1 ,, xL (7)
ai and bi like
 1T  t1T 
f L  x j   i G  ai , bi , x j  ,
L
j  1,., N . (4)    
i 1    and T   (8)
L 
T t LT 
Eq. (4) could be   Lm   N m

Table 5 Comparative performance statistics of ELM, ANN and GP in moment prediction

ELM ANN GP
2 2
RMSE R r RMSE R r RMSE R2 r
0.326023 0.9013 0.949345 0.8672 0.931215 0.376358 0.931347 0.326023 0.9013

Table 6 Comparative performance statistics of ELM, ANN and GP in rotation prediction

ELM ANN GP
RMSE R2 r RMSE R2 r RMSE R2 r
0.0183 0.9157 0.956927 0.023317 0.8607 0.927737 0.023314 0.8623 0.928585
 Moment-rotation estimation of steel rack connection using extreme learning machine
H is the hidden layer output matrix of SLFN with ith
column of H being the ith hidden node‘s output regarding to
the inputs x1,..., xN.
2.8.2 ELM principle
ELM has been defined as a SLFN with 𝐿 hidden
neurons to learn L distinct samples with 0 error. When a
number of hidden neurons L < the number of distinct
samples N, ELM has randomly assigned the characteristics
of hidden nodes and computed the output weights by
pseudo instead of H offering one small error ε > 0. The
hidden node characteristics of ELM as ai and bi couldn‘t be
tuned in train and is applied with random variables as the
thesis below:
Theorem 1: Let an SLFN with L additive or RBF hidden
nodes and an activation function g(x) which is infinitely
different in any interval of R be given (Liang et al. 2006).
Arbitrary L distinct input vectors {xi | xi  Rn, i = 1,..., L}
and { 𝑎𝑖 , 𝑏𝑖 }𝐿𝑖=1 have been randomly developed by any
continuous probability distribution, the hidden layer
outcome matrix is invertible with probability one, the
hidden layer output matrix H of the SLFN is invertible and
||Hβ ‒ T|| = 0.
Theorem 2: Given any small positive variable ε > 0 and
activating performance g(x): R → R, unlimited
discrepancies in any interval, there would be L ≤ N, then for
N arbitrary distinct input vectors {xi | xi  Rn, i = 1,..., L},
for any { 𝑎𝑖 , 𝑏𝑖 }𝐿𝑖=1 that are randomly developed
according to any continuous probable distributing ||HN×LβL×m
‒ TN×m|| < ε with probability one (Liang et al. 2006). Since
the hidden node characteristics of ELM couldn‘t be tuned
across the train or easily assigned with random variables,
Eq. (5) has become a linear system and the output weights
could be computed as (Khu et al. 2001)
  H T (9)
H+ is the Moore Penrose generalized inverse of hidden
layer output matrix H calculated by few models comprising
orthogonal projection, orthogonalization, repeating, and
Singular Value Decomposition (SVD). The orthogonal
projection has been applied if HTT is non-singular and H+
= (HTT) ‒ 1 HT. Because of the seeking and repeating,
orthogonalization and iterative approach are including few
curbs. The use of ELM has applied SVD to analyze the
Moore Penrose generalized inverse of H, due to its general
usage in all cases proving ELM as a batch learning
approach.
2.8 ANN
The multilayer feedforward network with BP is a very
common NN, precisely investigated and commonly applied
in more realms. Generally, one NN has consisted of three
layers: (1) one input layer; (2) one output layer; and (3) one
intermediate or hidden layer. The input vectors are 𝐷 ∈ 𝑅𝑛
and 𝐷 = (𝑋1 , 𝑋2 , … , 𝑋𝑛 )𝑇 ; the outputs of q neurons in the
hidden layer are 𝑍 = (𝑍1 , 𝑍2 , … , 𝑍𝑛 )𝑇 ; and the results of the
output layer are
m T
Y ∈ R , Y = (Y1 , Y2 , … , Yn )
431
wij and yj = weight and threshold between the input layer
and hidden layer
wjk and yk = weight and threshold between the hidden
layer and output layer
The outcomes of any neuron in one hidden layer and
output layer are
𝑛
𝑍𝑗 = 𝑓 𝑤𝑖𝑗 𝑋𝑖 − 𝜃𝑗 (10)
𝑖=1
𝑞
𝑌𝑘 = 𝑓 𝑤𝑘𝑗 𝑍𝑗 − 𝜃𝑘 (11)
𝑗 =1
𝑓 is transferring performance or mapping rule to sum
up the neuron‘s input to its output, and applies a proper
option of an instrument to nominate a nonlinearity into the
network designing. Another typically applied function is
sigmoid function as a monotonic increment ranging from
zero to one.
2.10 GP
GP as an evolutionary algorithm is according to the
natural choice and survival to approximate the equation
(Darwinian theories), ironically describing how the output
has related to input values, so it is about the primitive
population of randomly generated equations, obtained from
randomly input values combinations, random functions and
numbers, including arithmetic operators +, −, × , ÷ ,
mathematic parameters as (sin, cos, exp, log), and logical
comparison performances properly derived from
appropriate process conception. Therefore, the population
of potential solution(s) has been adjusted to the
evolutionary procedure and fitness (measuring the quality
of problem solving) of the evaluated evolved programs. In
the following, separate programs (best fitted to the data)
have been obtained from the primitive population, and then
these best fitted programs have been chosen to change the
part of information between the programs to make the best
ones by ‗crossover‘ and ‗mutation‘, imitating the real
reproduction procedure of nature. Accordingly, the process
of changing parts within the (best) programs is crossover,
and randomly changing programs to create the new ones is
mutation. Thereafter, those programs with a less fitting -
data have been discarded. This evolution process has been
iterated on the sequential generations to find the symbolic
explanations about data that scientifically has been
described to gain the knowledge about the process.
3. Results and discussion
3.1 Accuracy evaluation of the proposed
approaches
The predictive function of the methods has been
provided as RMSE, R2 and r
n
 (P  O )
i i
2
(12)
RMSE  i 1
n
432 Mahdi Shariati, Nguyen Thoi Trung, Karzan Wakil, Peyman Mehrabi, Maryam Safa and Majid Khorami

(a) (b) (c)

Fig. 4 the scatter plots of real and predicted variables in moment through: (a) ELM; (b) GP; (c) ANN

(a) (b) (c)

Fig. 5 the scatter plots of actual and predicted variables in rotation through: (a) ELM; (b) GP; (c) ANN

 n   n   n  ELM, confirmed by higher R2 variable. The number of

n   Oi  Pi     Oi     Pi  overestimated and underestimated maintained variables is
r  i 1   i 1   i 1  confined. Then, the predicted variables have obviously
 n 2  n (13)
   n  n  
2 2
enjoyed from high level precision.
 n  Oi    Oi     n  Pi 2    Pi  
 i 1  i 1    i 1  i 1  

4. ELM, ANN and GP performance comparing
2
 n 
 
  Oi  Oi  Pi  Pi  To define the positive features of ELM in an obvious
R 2   ni 1  (14) basis, the prediction accuracy of ELM has been juxtaposed

   
n
by the prediction accuracy of GP and ANN applied as a
 Oi  Oi   Pi  Pi
i 1 i 1
benchmark by using RMSE, r and R2. Tables 5-6 have
summarized the predicting accuracy for test data-set,
Pi and Oi = test and predict variables of steel rack because the training error is not a reliable index to predict
connection moment and rotation the potential of particular method. According to the
n = total test data number root means square error outcomes, ELM has drastically outperformed than GP and
(RMSE), ANN, in other word, based on RMSE analysis, ELM has
Coefficient of determination (R2) significantly provided better results than benchmark models
Pearson coefficient (r) (ANN and GP).

3.2 Performance evaluation of ELM

This section has focused on the outcomes of ELM steel
rack connection moment and rotation predictive models of
building as follows:
The accuracy of developed ELM in moment and
rotation predictive methods is illustrated in Fig. 4(a). The
accuracy of developed GP and ANN in moment predicting
methods is illustrated in Figs. 4(b)-(c). The accuracy of
developed ELM, GP and ANN in rotation predictive
methods is illustrated in Figs. 5(a)-(c).
Therefore, the majority of points have fallen along the
diagonal line for ELM predicting models. So, the prediction
outcomes have been aligned with the measured variables in
5. Conclusions
The current research has performed a systematic model
to create ELM steel rack connection moment and rotation
predictive model. The comparison of ELM with GP and
ANN has been used to verify the prediction accuracy,
thereafter, by using RMSE, r and R2, the results has
confirmed the superiority of ELM to GP and ANN. On the
other hand, ELM has few more demanding properties
distinguishing it from the classic (ordinary) gradient based
learning algorithms for feed forward NN. ELM is fast in
learning than the classic feed forward network learning
algorithms like BP, also, ELM has the efficiency of
 Moment-rotation estimation of steel rack connection using extreme learning machine
achieving the lowest train error and also weights‘ norm. The
current study has shed light on the computer science
applications in steel rack connections. Consequently, the
estimating procedure in moment and rotation has assisted
the construction section having a broad view in structure
behavior across the design, constructing and operations
resulting to more realistic and accurate design and
construction.
References
Abedini, M., Khlaghi, E.A., Mehrmashhadi, J., Mussa, M.H.,
Ansari, M. and Momeni, T. (2017), ―Evaluation of Concrete
Structures Reinforced with Fiber Reinforced Polymers Bars, A
Review‖, J. Asian Sci. Res., 7(5), 165-175.
Abedini, M., Mutalib, A.A., Mehrmashhadi, J., Raman, S.N.,
Alipour, R., Momeni, T. and Mussa, M.H. (2019), ―Large
Deflection Behavior Effect in Reinforced Concrete Columns
Exposed to Extreme Dynamic Loads‖.
Aghakhani, M., Suhatril, M., Mohammadhassani, M., Daie, M.
and Toghroli, A. (2015), ―A simple modification of homotopy
perturbation method for the solution of Blasius equation in
semi-infinite domains‖, Math. Problems Eng.
Andalib, Z., Kafi, M.A., Kheyroddin, A. and Bazzaz, M. (2014),
―Experimental investigation of the ductility and performance of
steel rings constructed from plates‖, J. Constr. Steel Res., 103,
77-88.
Andalib, Z., Kafi, M.A., Kheyroddin, A., Bazzaz, M. and
Momenzadeh, S. (2018), ―Numerical evaluation of ductility and
energy absorption of steel rings constructed from plates‖, Eng.
Struct., 169, 94-106.
Anicic, O., Jović, S., Skrijelj, H. and Nedić, B. (2017), ―Prediction
of laser cutting heat affected zone by extreme learning
machine‖, Optics Lasers Eng., 88, 1-4.
Annema, A.J., Hoen, K. and Wallinga, H. (1994), ―Precision
requirements for single-layer feedforward neural networks‖,
Proceedings of the Fourth International Conference on
Microelectronics for Neural Networks and Fuzzy Systems, pp.
145-151.
Bazzaz, M., Kheyroddin, A., Kafi, M.A. and Andalib, Z. (2011),
―Evaluating the performance of steel ring in special bracing
frame‖, Proceedings of the 6th International Conference of
Seismology and Earthquake Engineering.
Bazzaz, M., Kheyroddin, A., Kafi, M.A. and Andalib, Z. (2012),
―Modeling and Analysis of Steel Ring Devised in Off-centric
Braced Frame with the Goal of Improving Ductility of Bracing
Systems‖, Iran Scientific and Industrial Researches
Organization; Tehran, Iran.
Bazzaz, M., Kafi, M.A., Kheyroddin, A., Andalib, Z. and Esmaeili,
H. (2014), ―Evaluating the seismic performance of off-centre
bracing system with circular element in optimum place‖, Int. J.
Steel Struct., 14(2), 293-304.
Bazzaz, M., Andalib, Z., Kheyroddin, A. and Kafi, M.A. (2015a),
―Numerical comparison of the seismic performance of steel
rings in off-centre bracing system and diagonal bracing system‖,
Steel Compos. Struct., Int. J., 19(4), 917-937.
Bazzaz, M., Andalib, Z., Kafi, M.A. and Kheyroddin, A. (2015b),
―Evaluating the performance of OBS-CO in steel frames under
monotonic load‖, Earthq. Struct., Int. J., 8(3), 699-712.
Bazzaz, M., Darabi, M.K., Little, D.N. and Garg, N. (2018), ―A
straightforward procedure to characterize nonlinear viscoelastic
response of asphalt concrete at high temperatures‖,
Transportation research record. Transportation Research Board
96th Annual Meeting, 2672, 481-492.
Bobaru, F.L.O.R.I.N., Mehrmashadi, J., Chen, Z.I.G.U.A.N.G. and
433
Niazi, S.I.N.A. (2018), ―Intraply Fracture in Fiber-Reinforced
Composites, A Peridynamic Analysis‖, Proceedings of the
American Society for Composites—Thirty-Third Technical
Conference.
Chen, C., Shi, L., Shariati, M., Toghroli, A., Mohamad, E.T., Bui,
D.T. and Khorami, M. (2019), ―Behavior of steel storage pallet
racking connection-A review‖, Steel Compos. Compos., Int. J.,
30(5), 457-469.
Davoodnabi, S.M., Mirhosseini, S.M. and Shariati, M. (2019),
―Behavior of steel-concrete composite beam using angle shear
connectors at fire condition‖, Steel Compos. Compos., Int. J.,
30(2), 141-147.
Gilbert, B.P. and Rasmussen, K.J. (2009), ―Experimental test on
steel storage rack components‖, TUoS School of Civil
Engineering, Australia (Ed.), School of Civil Engineering, The
University of Sydney, Australia.
Gilbert, B.P., Rasmussen, K.J., Baldassino, N., Cudini, T. and
Rovere, L. (2012), ―Determining the transverse shear stiffness
of steel storage rack upright frames‖, J. Constr. Steel Res., 78,
107-116.
Hamidian, M., Shariati, A., Khanouki, M.A., Sinaei, H., Toghroli,
A. and Nouri, K. (2012), ―Application of Schmidt rebound
hammer and ultrasonic pulse velocity techniques for structural
health monitoring‖, Scientific Research and Essays, 7(21),
1997-2001.
Heydari, A. and Shariati, M. (2018), ―Buckling analysis of tapered
BDFGM nano-beam under variable axial compression resting
on elastic medium‖, Struct. Eng. Mech., Int. J., 66(6), 737-748.
Ho-Huu, V., Vo-Duy, T., Duong-Gia, D. and Nguyen-Thoi, T.
(2018), ―An efficient procedure for lightweight optimal design
of composite laminated beams‖, Steel Compos. Struct., Int. J.,
27(3), 297-310.
Hosseinpour, E., Baharom, S., Badaruzzaman, W.H.W., Shariati,
M. and Jalali, A. (2018), ―Direct shear behavior of concrete
filled hollow steel tube shear connector for slim-floor steel
beams‖, Steel Compos. Struct., Int. J., 26(4), 485-499.
Huang, G.B., Zhu, Q.Y. and Siew, C.K. (2004), ―Extreme learning
machine, a new learning scheme of feedforward neural
networks‖, Proceedings of 2004 IEEE International Joint
Conference on Neural Networks.
Huang, G.B., Chen, L. and Siew, C.K. (2006a), ―Universal
approximation using incremental constructive feedforward
networks with random hidden nodes‖, Neural Networks, IEEE
Transactions on, 17(4), 879-892.
Huang, G.B., Zhu, Q.Y. and Siew, C.K. (2006b), ―Extreme
learning machine, theory and applications‖, Neurocomputing,
70(1), 489-501.
Huang, G.B., Zhu, Q.Y. and Siew, C.K. (2006c), ―Real-time
learning capability of neural networks‖, Neural Networks, IEEE
Transactions on, 17(4), 863-878.
Ismail, M., Shariati, M., Awal, A., Chiong, C., Chahnasir, E.,
Porbar, A., Heydari, A. and Khorami, M. (2018), ―Strengthening
of bolted shear joints in industrialized ferrocement
construction‖, Steel Compos. Struct., Int. J., 28(6), 681-690.
Kheyroddin, A., Bazzaz, M., Navan, Y.S. and Bazzaz, M. (2008),
―The complication and weakness of construction of steel
structure‖, Proceedings of the 14th International Civil
Engineering Student Conference, Semnan, Iran, 14, 1-9.
Khorami, M., Alvansazyazdi, M., Shariati, M., Zandi, Y., Jalali, A.
and Tahir, M. (2017a), ―Seismic performance evaluation of
buckling restrained braced frames (BRBF) using incremental
nonlinear dynamic analysis method (IDA)‖, Eartq. Struct., Int.
J., 13(6), 531-538.
Khorami, M., Khorami, M., Motahar, H., Alvansazyazdi, M.,
Shariati, M., Jalali, A. and Tahir, M.M. (2017b), ―Evaluation of
the seismic performance of special moment frames using
incremental nonlinear dynamic analysis‖, Struct. Eng. Mech.,
434 Mahdi Shariati, Nguyen Thoi Trung, Karzan Wakil, Peyman Mehrabi, Maryam Safa and Majid Khorami
Int. J., 63(2), 259-268.
Khu, S.T., Liong, S.Y., Babovic, V., Madsen, H. and Muttil, N.
(2001), ―Genetic Programming and Its Application in Real‐Time
Runoff Forecasting‖, JAWRA Journal of the American Water
Resources Association, 37(2), 439-451.
Liang, N.Y., Huang, G.B., Saratchandran, P. and Sundararajan, N.
(2006), ―A fast and accurate online sequential learning
algorithm for feedforward networks‖, Neural Networks, IEEE
Transactions on, 17(6), 1411-1423.
Luo, Z., Sinaei, H., Ibrahim, Z., Shariati, M., Jumaat, Z., Wakil,
K., Pham, B.T., Mohamad, E.T. and Khorami, M. (2019),
―Computational and experimental analysis of beam to column
joints reinforced with CFRP plates‖, Steel Compos. Struct., Int.
J., 30(3), 271-280.
Mansouri, I., Safa, M., Ibrahim, Z., Kisi, O., Tahir, M.M.,
Baharom, S. and Azimi, M. (2016), ―Strength prediction of
rotary brace damper using MLR and MARS‖, Struct. Eng.
Mech., Int. J., 60(3), 471-488.
Mansouri, I., Shariati, M., Safa, M., Ibrahim, Z., Tahir, M.M. and
Petković, D. (2017), ―Analysis of influential factors for
predicting the shear strength of a V-shaped angle shear
connector in composite beams using an adaptive neuro-fuzzy
technique‖, Journal of Intelligent Manufacturing, 1-11.
Mehrmashhadi, J., Chen, Z., Zhao, J. and Bobaru, F. (2019a), ―A
Stochastically Homogenized Peridynamic Model for Intraply
Fracture in Fiber-Reinforced Composites‖, 31.
Mehrmashhadi, J., Tang, Y., Zhao, X., Xu, Z., Pan, J.J., Van Le, Q.
and Bobaru, F. (2019b), ―The Effect of Solder Joint
Microstructure on the Drop Test Failure—A Peridynamic
Analysis‖, IEEE Transactions on Components, Packaging and
Manufacturing Technology, 9(1), 58-71.
Mohammadhassani, M., Nezamabadi-Pour, H., Jumaat, M.,
Jameel, M., Hakim, S.J.S. and Zargar, M. (2013), ―Application
of the ANFIS model in deflection prediction of concrete deep
beam‖, Struct. Eng. Mech., Int. J., 45(3), 319-332.
Mohammadhassani, M., Suhatril, M., Shariati, M. and Ghanbari,
F. (2014), ―Ductility and strength assessment of HSC beams
with varying of tensile reinforcement ratios‖, Struct. Eng.
Mech., Int. J., 48(6), 833-848.
Mohammadhassani, M., Saleh, A., Suhatril, M. and Safa, M.
(2015), ―Fuzzy modelling approach for shear strength prediction
of RC deep beams‖, Smart Struct. Syst., Int. J., 16(3), 497-519.
Nasrollahi, S., Maleki, S., Shariati, M., Marto, A. and Khorami,
M. (2018), ―Investigation of pipe shear connectors using push
out test‖, Steel Compos. Struct., Int. J., 27(5), 537-543.
Paknahad, M., Shariati, M., Sedghi, Y., Bazzaz, M. and Khorami,
M. (2018), ―Shear capacity equation for channel shear
connectors in steel-concrete composite beams‖, Steel Compos.
Struct., Int. J., 28(4), 483-494.
Sadeghi Chahnasir, E., Zandi, Y., Shariati, M., Dehghani, E.,
Toghroli, A., Mohamed, E.T., Shariati, A., Safa, M., Wakil, K.
and Khorami, M. (2018), ―Application of support vector
machine with firefly algorithm for investigation of the factors
affecting the shear strength of angle shear connectors‖, Smart
Struct. Syst., Int. J., 22(4), 413-424.
Safa, M., Shariati, M., Ibrahim, Z., Toghroli, A., Baharom, S.B.,
Nor, N.M. and Petkovic, D. (2016), ―Potential of adaptive neuro
fuzzy inference system for evaluating the factors affecting steel-
concrete composite beam‘s shear strength‖, Steel Compos.
Struct., Int. J., 21(3), 679-688.
Sajedi, F. and Shariati, M. (2019), ―Behavior study of NC and
HSC RCCs confined by GRP casing and CFRP wrapping‖, Steel
Compos. Struct., Int. J., 30(5), 417-432.
Sedghi, Y., Zandi, Y., Shariati, M., Ahmadi, E., Moghimi Azar, V.,
Toghroli, A., Safa, M., Tonnizam Mohamad, E., Khorami, M.
and Wakil, K. (2018a), ―Application of ANFIS technique on
performance of C and L shaped angle shear connectors‖, Smart
Struct. Syst., Int. J., 22(3), 335-340.
Shah, S.N.R., Sulong, N.R., Shariati, M. and Jumaat, M.Z (2015),
―Steel Rack Connections, Identification of Most Influential
Factors and a Comparison of Stiffness Design Methods‖, PloS
one, 10(10), e0139422.
Shah, S.N.R., Sulong, N.R., Jumaat, M.Z. and Shariati, M.
(2016a), ―State-of-the-art review on the design and performance
of steel pallet rack connections‖, Eng. Fail. Anal., 66, 240-258.
Shah, S.N.R., Sulong, N.R., Shariati, M., Khan, R. and Jumaat,
M.Z. (2016b), ―Behavior of steel pallet rack beam-to-column
connections at elevated temperatures‖, Thin-Wall. Struct., 106,
471-483.
Shah, S.N.R., Sulong, N.R., Khan, R., Jumaat, M.Z. and Shariati,
M. (2016c), ―Behavior of industrial steel rack connections‖,
Mech. Syst. Signal Process., 70-71, 725-740.
Shariat, M., Shariati, M., Madadi, A. and Wakil, K. (2018),
―Computational Lagrangian Multiplier Method by using for
optimization and sensitivity analysis of rectangular reinforced
concrete beams‖, Steel Compos. Struct., Int. J., 29(2), 243-256.
Shariati, M., Sulong, H., Maleki, S. and Kh, M. (2010),
―Experimental and analytical study on channel shear connectors
in light weight aggregate concrete‖, Proceedings of the 4th
International Conference on Steel & Composite Structures,
Sydney, Australia, July.
Shariati, M., Sulong, H., Suhatril, M., Shariati, A., Arabnejad
Khanouki, M.M. and Sinaei, H. (2012), ―Fatigue energy
dissipation and failure analysis of channel shear connector
embedded in the lightweight aggregate concrete in composite
bridge girders‖, Proceedings of the Fifth International
Conference on Engineering Failure Analysis, The Hague, The
Netherlands, July.
Shariati, M., Sulong, N.R., Shariati, A. and Khanouki, M.A.
(2015), ―Behavior of V-shaped angle shear connectors,
experimental and parametric study‖, Mater. Struct., 1-18.
Shariati, M., Toghroli, A., Jalali A. and Ibrahim, Z. (2017),
―Assessment of stiffened angle shear connector under
monotonic and fully reversed cyclic loading‖, Proceedings of
the Fifth International Conference on Advances in Civil,
Structural and Mechanical Engineering - CSM 2017, Zurich,
Switzerland.
Shariati, M., Tahir, M.M., Wee, T.C., Shah, S.N.R., Jalali, A.,
Abdullahi, M.A.M. and Khorami, M., (2018), ―Experimental
investigations on monotonic and cyclic behavior of steel pallet
rack connections‖, Eng. Fail. Anal., 85, 149-166.
Singh, T.N., Gulati, A., Dontha, L. and Bhardwaj, V. (2008),
―Evaluating cut slope failure by numerical analysis—a case
study‖, Nat. Hazards, 47(2), 263.
Singh, R., Kainthola, A. and Singh, T.N. (2012), ―Estimation of
elastic constant of rocks using an ANFIS approach‖, Appl. Soft
Comput., 12(1), 40-45.
Toghroli, A. (2015), ―Applications of the ANFIS and LR Models
in the Prediction of Shear Connection in Composite Beams‖,
Jabatan Kejuruteraan Awam, Fakulti Kejuruteraan, Universiti
Malaya.
Toghroli, A., Mohammadhassani, M., Suhatril, M., Shariati, M.
and Ibrahim, Z. (2014), ―Prediction of shear capacity of channel
shear connectors using the ANFIS model‖, Steel Compos.
Struct., Int. J., 17(5), 623-639.
Toghroli, A., Suhatril, M., Ibrahim, Z., Safa, M., Shariati, M. and
Shamshirband, S. (2016), ―Potential of soft computing approach
for evaluating the factors affecting the capacity of steel–
concrete composite beam‖, J. Intel. Manuf., 1-9.
Toghroli, A., Darvishmoghaddam, E., Zandi, Y., Parvan, M., Safa,
M., Abdullahi, M.A.M., Heydari, A., Wakil, K., Gebreel, S.A.
and Khorami, M. (2018), ―Evaluation of the parameters
affecting the Schmidt rebound hammer reading using ANFIS
method‖, Comput. Concrete, Int. J., 21(5), 525-530.
 Moment-rotation estimation of steel rack connection using extreme learning machine 435

Vo-Duy, T., Duong-Gia, D., Ho-Huu, V., Vu-Do, H.C. and

Nguyen-Thoi, T. (2017), ―Multi-objective optimization of
laminated composite beam structures using NSGA-II
algorithm‖, Compos. Struct., 168, 498-509.
Vo-Duy, T., Ho-Huu, V. and Nguyen-Thoi, T. (2018), ―Free
vibration analysis of laminated FG-CNT reinforced composite
beams using finite element method‖, Front. Struct. Civil Eng.,
13(2), 324-336.
Wang, X. and Han, M. (2014), ―Online sequential extreme
learning machine with kernels for nonstationary time series
prediction‖, Neurocomputing, 145, 90-97.
Wei, X., Shariati, M., Zandi, Y., Pei, S., Jin, Z., Gharachurlu, S.,
Abdullahi, M.M., Tahir, M.M. and Khorami, M. (2018),
―Distribution of shear force in perforated shear connectors‖,
Steel Compos. Struct., Int. J., 27(3), 389-399.
Xie, Q., Sinaei, H., Shariati, M., Khorami, M., Mohamad, E.T. and
Bui, D.T. (2019), ―An experimental study on the effect of CFRP
on behavior of reinforce concrete beam column connections‖,
Steel Compos. Struct., Int. J., 30(5), 433-441.
Yu, Q., Miche, Y., Séverin, E. and Lendasse, A. (2014),
―Bankruptcy prediction using extreme learning machine and
financial expertise‖, Neurocomputing, 128, 296-302.
Zandi, Y., Shariati, M., Marto, A., Wei, X., Karaca, Z., Dao, D.,
Toghroli, A., Hashemi, M.H., Sedghi, Y., Wakil, K. and
Khorami, M. (2018), ―Computational investigation of the
comparative analysis of cylindrical barns subjected to
earthquake‖, Steel Compos. Struct., Int. J., 28(4), 439-447.

CC